Modeling interconnects over silicon
While I enjoyed reading the hilarious comments in the commodity thread,
here is a technical problem that has challenged me from time to time and
that I wish to find more insightful inputs and advice here.
We are modeling long traces over the top of silicon. The distance is small
enough to justify the inclusion of coupling between this trace and the top
metal layers of silicon. The challenge here is how to handle the dense
power grid in the 3D EM model.
Option 1, a simple way is to treat the entire power grid as a plane and use
it as a reference plane for the trace. There has been some reasonable
correlation with this approach. The potential sources of inaccuracies are
obvious too, including the discontinuity between power and ground, and the
orientation difference between the grid and the trace. Has anyone done
similar models before and could please comment on the concerns of
Option 2, a hard way is to import these GDS layers into the trace database,
and model them together with a EM tool. There are many layers of
complexity: (1) The size of the problem is too big for typical 3D FEM based
solver. (2) You might have to handle the multiple power and ground stripes
in the grid layer by connecting decaps at both silicon and board level,
unless simply treating every stripe as ground is an OK approximation.
Could anyone share some insights or experiences on this modeling option? Is
there any known MOM-based EM tool or PG grid tool like Redhawk that can
handle such a model? Any papers or research study that you can suggest?
BTW, I have done some quick literature survey, and it seems there is some
limited research on the FDTD solver for power-grid, some FEM eigenvalue
solver for grid resonance, etc... Somehow, I am not able to find relevant
studies on this subject.
phillypham 5 years 1 month 12 days